Andy, Thank you for the comments!
On Sun, Nov 23, 2008 at 2:42 PM, Andy Turner <[EMAIL PROTECTED]> wrote: > Hi Simon, (geowanking list) > > I have been doing some research into road accident incidence in the UK. I > started this work back in about 2000 and it has been on the back burner for a > while. > > It is very difficult to know what is a sensible denominator for a rate or a > comparison to look at incidence of road accidents. > > You can try to build up an explanatory model of the variation, but I found > that nothing really works because the theoretically desirable data is not > available. Both intrinsic risk of a fatal road accident and exposure to risk > are interrelated and change over time. No argument that doing this with the scientific rigor is hard. However, I did not have any assumptions about the data, and I just wrote about the obvious things that seemed to stand out to my uneducated mind. I want to follow through on the rural/urban theory by looking at the crash data by county and by correlating them with the degree of ruralness of each county, which should yield a slightly more robust comparison. > I can argue that knowing where all the vehicles are at any time to a high > level of spatial and temporal detail should help, but sadly I don't expect > that data to become available for this research for some time yet... > > What use is an analysis at state resolution, or indeed any resolution? What > spatial resolution do the incidence data have? I think looking at states and counties makes somewhat more sense in the US than in the UK, as the regional/cultural differences are more significant here, while infrastructure level and driving patterns are broadly uniform across both countries. For example, Utah has very few drunk driving accidents, and I'm sure that's not because they have very good driver's ed. So I don't think I'm seeing completely false effects, as, for example, this report: http://www-nrd.nhtsa.dot.gov/Pubs/810968.PDF also mentions the Southwest region as being more dangerous for pedestrians. Most of the incident data for 2005-7 (93%) are geolocated precisely. I don't know how accurate the coordinates are, but I have not seen any clearly misplaced points yet. For rigorous analysis, it would be vital to look at non-fatal accidents too, but this data is not fully available yet. I'm currently trying to make sense of the California data for all accidents that is fully available, but it's geolocated by intersection, not by coordinates, so it's hard to work with. The GES program (http://www.nhtsa.dot.gov/portal/site/nhtsa/menuitem.0efe59a360fbaad24ec86e10dba046a0/) collects a representative subset of all crash data for non-geographic analysis, but since it's only a subset and since the data lack coordinates, I can't do much with them. As for why I wanted to look at full-resolution data, I was just curious to see if there any obvious accident clusters in my neighborhood, and I think other people will want to see this too. Of course, this is not something that should be used for urban planning. > I found it interesting to look at change over time (for one or more years) at > as high a level of spatial resolution as the data permits. In doing this it > is useful to use some distance weighted generalisation to make the patterns > visible while examining large regions. The first thing you might notice are > new roads, but also it is possible to see geographical patterns and wonder... I decided to not plot the data for the three years separately, as there does not seem to be enough variation. Taking the data with, say, five years' interval would have been more useful, but it's too early to do that. Good point about new roads - this would need to be taken into account. > Yes, and when I observed new roads and roads moving I wondered if this was > captured in available digital map data. At the time the data did not exist, > but maybe it is getting there now :-) > > I would like to have a crowd sourcing application that allows the public to > say where they think is dangerous and where they have had near misses. This > might be very useful data for such analysis. I'm generally a fan of crowdsourcing, but I don't think in this case it would be more useful than simply collecting and geolocating all police crash reports. A near miss is very subjective, and the selection of people who would report them would be biased. However, reporting inconvenient or dangerous road conditions to local authorities sounds like a useful application. Thanks, Simon _______________________________________________ Geowanking mailing list [email protected] http://geowanking.org/mailman/listinfo/geowanking_geowanking.org
